By eliminating user-centric biases and adopting a purely item-focused approach, it is possible to achieve ethical and effective ranking systems—ensuring fairness, resilience, and compliance with regulations on responsible AI.

Ranking systems are essential in online platforms, shaping user experiences and influencing product visibility and sales. However, traditional user-centric ranking systems, which assign reputation scores to users, raise ethical concerns like discrimination, bias, and vulnerability to spam and bribery. In a recent study, conducted with Guilherme Ramos and Mirko Marras, we propose an innovative, ethically grounded solution to the item ranking problem.

Ethical issues in user-centric rankings

User-centric ranking systems rely on reputation scores assigned to users based on their ratings. While effective in combating spam and bribery, this approach:

1. Risks bias and discrimination due to reliance on user attributes.
2. Contravenes principles in regulations like the EU AI Act, which prohibits social scoring.
3. Misuses personal data for deriving reputation scores.

Our solution: a user-agnostic ranking system

The study introduces a user-agnostic ranking system (UARS) that evaluates items based solely on their ratings, filtering out anomalies through an iterative process. This paradigm:

1. Removes ratings significantly deviating from the mean.
2. Avoids reliance on user reputation, ensuring fairness and compliance with ethical standards.

Algorithm

UARS iteratively calculates the mean and standard deviation of ratings for each item, discarding outlier ratings until convergence. This results in rankings that are:

• Efficient. Computationally faster than user-centric models.
• Fair. Independent of user attributes, thus non-discriminatory.
• Resilient. Robust against spam and bribery attempts.

Experimental validation

The proposed system was tested on three datasets: MovieLens-100k, MovieLens-1M, and Amazon Musical Instruments. Our main findings include:

1. Efficiency. UARS demonstrated computational speed up to five times faster than user-centric systems.
2. Effectiveness. Rankings closely aligned with arithmetic averages while being more resistant to manipulation.
3. Robustness. High resilience to spamming, as measured by Kendall’s Tau, maintaining ranking stability even with spam-injected datasets.
4. Bribery resistance. UARS significantly reduced the financial advantage of bribery, outperforming existing systems in deterring fraudulent practices.

Ethical implications and future directions

UARS represents a significant step towards ethical AI in ranking systems, fostering trust and fairness in online platforms. Future research aims to:

1. Investigate advanced bribing strategies and their countermeasures.
2. Integrate machine learning for dynamic adaptability to evolving user behaviors.
3. Expand testing to more diverse datasets and real-world applications.